Download Locating hydrothermal acoustic sources at Old Faithful Geyser using

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Seismic anisotropy wikipedia , lookup

Global Energy and Water Cycle Experiment wikipedia , lookup

Seismic communication wikipedia , lookup

Geophysical MASINT wikipedia , lookup

Earthquake engineering wikipedia , lookup

Surface wave inversion wikipedia , lookup

Reflection seismology wikipedia , lookup

Seismic inversion wikipedia , lookup

Seismometer wikipedia , lookup

Earthscope wikipedia , lookup

Transcript
Geophysical Journal International
Geophys. J. Int. (2011) 187, 385–393
doi: 10.1111/j.1365-246X.2011.05147.x
Locating hydrothermal acoustic sources at Old Faithful Geyser using
Matched Field Processing
E. Cros,1 P. Roux,2 J. Vandemeulebrouck1 and S. Kedar3
1 ISTerre,
CNRS UMR5275, Université de Savoie, Campus Scientifique, 73376 Le Bourget-du-Lac Cedex, France. E-mail: [email protected]
CNRS UMR5275, Université Joseph Fourier, Maison des Géosciences, BP 53, 38041 Grenoble Cedex 9, France
3 Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA 91109, USA
2 ISTerre,
Accepted 2011 July 10. Received 2011 July 1; in original form 2011 April 27
SUMMARY
In 1992, a large and dense array of geophones was placed around the geyser vent of Old
Faithful, in the Yellowstone National Park, to determine the origin of the seismic hydrothermal
noise recorded at the surface of the geyser and to understand its dynamics. Old Faithful
Geyser (OFG) is a small-scale hydrothermal system where a two-phase flow mixture erupts
every 40 to 100 min in a high continuous vertical jet. Using Matched Field Processing (MFP)
techniques on 10-min-long signal, we localize the source of the seismic pulses recorded at
the surface of the geyser. Several MFP approaches are compared in this study, the frequencyincoherent and frequency-coherent approach, as well as the linear Bartlett processing and the
non-linear Minimum Variance Distorsionless Response (MVDR) processing. The different
MFP techniques used give the same source position with better focalization in the case of the
MVDR processing. The retrieved source position corresponds to the geyser conduit at a depth
of 12 m and the localization is in good agreement with in situ measurements made at Old
Faithful in past studies.
1 I N T RO D U C T I O N
Old Faithful Geyser (OFG) is located in the Upper Geyser Basin
(UGB) in the Yellowstone National Park. The basin is approximately 3.2 × 0.8 km and is on the periphery of the Mallard Lake
resurgent dome within the Yellowstone Caldera. The caldera was
formed 640 000 yr ago by a giant eruption and measures 80 × 50 km.
The presence of a complex magmatic reservoir system beneath the
caldera (Fournier 1989; Miller & Smith 1999; Husen et al. 2004)
delivers the heat that maintains the hot springs, geysers and mud
pots on the different basins. The heat flux density calculated with a
river chloride inventory method is estimated at 2000 mW m−2 over
the 2900 km2 corresponding to the caldera area (Fournier 1989).
The Yellowstone thermal water is mainly meteoric in origin with
magmatic contribution less than a few per cent (Fournier 1989).
The edifice of OFG is essentially conical with a diameter of
60 m, and is characterized by a 4-m-high geyserite vent concretion,
with an opening of 2 m × 1 m, and an irregular, elongated fissurelike conduit (Hutchinson et al. 1997). It is one of the most studied
geyser in the world because of its regularity and the short interval
between two eruptions which makes its study very convenient. The
time interval between two eruptions follows a bimodal distribution,
between 40 and 100 min, with a principal mode centred on 80 min.
The eruption is characterized by a continuous vertical jet of water
and steam at a height of 30–50 m lasting for 1–6 min for a total
discharge of water between 14 000 and 32 000 l.
C
2011 The Authors
C 2011 RAS
Geophysical Journal International Most of the studies on Old Faithful focalize on the evolution of
the time interval between two eruptions (Rinehart 1969; Hurwitz
et al. 2008), on the characterization of the seismic signals recorded
around the vent (Kieffer 1984; Kedar et al. 1996, 1998) or on
the working-out of a dynamic model of the cycle (Kieffer 1984;
Hutchinson et al. 1997).
Kieffer (1984) was the first to give an elaborate description of
the OFG behaviour, including its seismicity and thermodynamics,
based on the data collected by Birch and Kennedy in 1948. She
established a model of the rise of water in the conduit before an
eruption (Fig. 1) and considered that the collapse of steam bubbles which cool in the upper part of the water column, is a major
physical process that transfers latent heat to the water column, and
produces impulsive acoustic events, which composed the seismic
signal recorded at the surface. Hutchinson et al. (1997), using pressure probes and a small video camera lowered in the conduit, were
able to observe different hydrodynamic processes occurring in the
conduit, such as boiling and cavitation, but also superheated steam
expansion, and exsolution of incondensable gas that they proposed
to be CO2 . From 1991 to 1994, Kedar and colleagues conducted
several seismic surveys at the surface simultaneously with pressure
and temperature measurements in the OFG conduit (Kedar et al.
1996). During their experiment, an array of 96 short period vertical
geophones, and several broad-band sensors were placed around the
geyser vent. At the surface they recorded a quasi-harmonic seismic
signal composed of the succession of very impulsive events. They
385
GJI Seismology
Key words: Hydrothermal systems; Volcano seismology; Wave propagation; North America.
386
E. Cros et al.
Figure 1. Water level in the conduit of Old Faithful before an eruption
reported by Birch & Kennedy (1972), from an adaptation of Kieffer (1984).
observed that the tremor intensity is modulated by the variations
in the conduit shape during water rise (Kedar et al. 1998). The
impulsive events composing the signal recorded at the surface reverberate in a soft shallow layer and are not generated by resonance
in the water column, as assumed by Kieffer (1984). Finally, pressure
measurements in the conduit confirmed that the individual seismic
pulses are generated by the collapse of bubbles.
The goal of this paper is to revisit the data recorded at OFG
by Kedar and his colleagues with the dense array of geophones
to check how acoustic source localization techniques derived from
ocean acoustics, namely Matched Field Processing (MFP), can be
used to localize the cavitation events recognized by Kieffer and
Kedar in the conduit during the cycle.
MFP is a well-established passive technique used to track submarines or marine mammals in the ocean, to differentiate animals
or to understand their behaviour (Thode et al. 2000). MFP was recently tested with success on a seismic array of 10 sensors deployed
on hydrothermal systems, exhibiting the dominant acoustic source
below the array (Legaz et al. 2009; Vandemeulebrouck et al. 2010).
In this case, the seismic array is six times denser than during the
aforementioned experiments on hydrothermal systems. Moreover,
the seismic sources are shallow and can be associated with in situ
measurements that provide additional constrains to this study.
2 D ATA
The network deployed by Kedar consists of 96 vertical 1 Hz geophones spread on a tight grid around the vent of the geyser (Fig. 2).
The 96 geophones originally recorded signal during several eruptions with a sampling frequency of 250 Hz, but only 10 min of signal
was readily available to be processed in this study. Nevertheless, this
10-min interval is associated with a stable stage of the geyser cycle,
when the water level is slowly rising in the conduit approximately
20 min before the eruption (Kedar et al. 1998, see Fig. 1). From
the array data, a map of the seismic intensity radiated at the surface around the geyser was calculated at different periods during
the cycle by Kedar (1996), and the results showed no significant
behaviour. Several hammer shots were also performed to complete
the study, to look at the difference between the excitation of the
medium by the hammer shot and by the seismic natural impulsive
sources.
Finally, the analysis of single seismic events recorded on the
geophones with a simultaneous measurement of the water pressure
in the water column at different depths was performed by Kedar
(1996) and Kedar et al. (1996, 1998). The fact that the record of a
pressure pulse near the top surface is followed by the record of an
impulsive event on the geophones clearly indicates that these events
are generated by bubble collapses in the water column.
Thus, the seismic signal recorded in this study is mainly composed of impulsive events (Figs 3a and c), with a duration in the
order of 0.2 s and with an approximate rate of 100 events per
minute (Fig. 3c). During an eruption, it was observed that the number of events before an eruption follows an asymptotical law (Kedar
1996), corresponding to the rise of the water level in the conduit.
The 10-min-long record processed in this study is mainly stable in
amplitude and does not show evidence of an eruption (Fig. 3a). It
actually corresponds to a period of approximately 20 min before
an eruption. The frequency content of the signal is large with two
modes, the first one, the most energetic, between 10 and 40 Hz and
the second one, between 50 and 65 Hz (Fig. 3b).
3 METHOD
3.1 Presentation of the MFP techniques
In volcano seismology, source localization is generally performed
on a series of single events of the same type (Very Long Period, Long
Period, Volcano-Tectonic and Tremor). Historically, time-picking
of arrival time has been performed on impulsive volcano-tectonics
events. When this method cannot be used, several other methods
exist to localize seismic events recorded on volcanoes. Among these
methods, cross-correlation technique permits to determine the time
delays between pairs of station and to compare these delays with
theoretical ones associated to a point source. This technique was
applied to localize Long Period events on Mt Etna (De Barros et al.
2009). The source positions retrieved were in agreement with the
localization given by time reversal on these same events (O’Brien
et al. 2011). The estimation of the slowness vector has also been
applied on volcanic signals of different types to locate their origin
(Almendros et al. 2001; Métaxian et al. 2002), as well as on the
subduction zone in the Cascades (La Rocca et al. 2010) to retrieve
the location of the tremor sources.
Similarly, one can retrieve the source location by looking at the
spatial amplitude distribution for several types of events recorded
across a network and comparing it with theoretical amplitude decay calculated for a given point source location. Assuming the
type of waves considered, that is, body waves or surface waves,
we can retrieve the source location which best fits the data (Aki
& Ferrazzini 2000; Battaglia & Aki 2003). The method was
C 2011 The Authors, GJI, 187, 385–393
C 2011 RAS
Geophysical Journal International Old Faithful viewed through array processing
387
Figure 2. (a) Shaded relief map of Old Faithful area. Coordinates are indicated in (m UTM). (b) Topographic map and location of the 96 vertical 1 Hz
geophones around the vent of Old Faithful Geyser. Elevation contour interval is 1 m. The red square corresponds to the grid where the MFP was processed.
The reference coordinates of the geyser vent are X = 513 672.51 m UTM, Y = 4 923 032.42 m UTM and Z = 2240 m.
successfully applied on rockfalls but faced difficulties when considering volcano-tectonic events occurring below the summit of
the volcano and below the sea level due to the complexity of the
amplitude distribution in this region.
Another technique using a set of similar earthquakes or Long Period events, called multiplets, consists in determining the difference
in origin times between each pair of events in the multiplet. Localization is then performed by minimizing the residuals between the
time delays between two events and theoretical time delays computed after relative relocation (Got et al. 1994; Battaglia et al. 2003).
Finally, the semblance method was used on tremors generated by a
volcanic eruption to follow the migration of the seismic activity between two potential sources on Izu-Oshima Island volcano in Japan
(Furumoto et al. 1990).
In geothermal areas, the seismic signal recorded at the surface
of the hydrothermal system is composed of randomly distributed
C
2011 The Authors, GJI, 187, 385–393
C 2011 RAS
Geophysical Journal International impulsive events related to bubble collapse (Kedar et al. 1998; Legaz
et al. 2009; Vandemeulebrouck et al. 2010; Ichihara & Nishimura
2011). In the case of the present data, the impulsive events often overlap and present very different signal-to-noise ratio on the
geophone array at the surface. This makes time-picking algorithms
inefficient to identify and relocalize each event. Furthermore, the
high rate of events (100 per minute, see Fig. 3c) would make an
event-by-event relocalization very time consuming. In this perspective, the advantage of the MFP technique is to build up a probability
of presence of the dominant acoustic source on a selected time window of the recorded signals. As a matter of fact, the goal of MFP is
to stack the events on a time interval T to provide a robust relative
phase measurement on the whole array. In other words, under the
approximation that most of the bubble collapses in the time window
T come from the same area (within the half-wavelength), the MFP
capitalizes on the phase coherence of these events recorded on the
388
E. Cros et al.
Figure 3. A 10-min-long record at geophone 90. (a) 10 min record. (b) Average amplitude spectrum calculated over the 96 geophones. (c) Zoom between 300
and 310 s.
array. Thus, MFP cumulates the advantage of (1) a better signal-tonoise ratio through the stacking of events in the time interval T and
(2) an automatic procedure to localize the dominant seismic source
as a function of time for long recordings.
Historically, MFP is a localization technique commonly used in
ocean acoustics that starts to be used on hydrothermal systems.
This array processing method is a generalization of beamforming
techniques in the sense that it basically compares phase delays of
forward modelling solutions of the wave equation to acquired data.
More precisely, MFP consists of placing a test source at each point
of a 3-D search grid, computing the acoustic field at all elements
of the array and then matching this modelled field with the data.
The match is maximum when a point source of the search grid is
colocated with the true point source. The result of the processing is
a probability map of the source position.
There exist different ways to match the modelled field to the data.
The linear method, called Bartlett MFP, performs a correlation between the data and the model. The non-linear method, in our case the
Minimum-Variance Distorsionless Response (MVDR), computes a
maximum-likelihood type minimization between the data and the
model. Compared to Bartlett, the MVDR technique improves the
resolution of the MFP output but it requires both a good signalto-noise ratio on the recorded data and a propagation model that
perfectly adjusts to the data (Jensen et al. 1995). On the other hand,
the Bartlett MFP gives a robust solution, even for low signal-tonoise ratio, with a spatial resolution that is limited to the acoustic
wavelength according to diffraction laws.
For both linear and non-linear MFP algorithms, the processing is
performed in the frequency domain as follows:
First, the Cross-Spectral Density Matrix (CSDM) K is calculated
as
∗
K =d ·d ,
(1)
= [d 1f , df 2 , . . . , dNf ] defined as the acoustic signal at frequency
with d
f recorded on a geophone i (i varying from 1 to N geophones). The
star indicates the complex conjugate transpose operation.
Second, a model-based replica vector dm (f , ai ) is defined at frequency f as the modelled field from a candidate source position to
the array elements, with ai being the vector corresponding to the absolute distance between the source candidate position and geophone
i of the array. In our case, the propagation model corresponds to the
free-space medium which means that the replica vector is expressed
by
1
−2πi f ai
.
(2)
exp
dm ( f, ai ) =
4πai
c
In eq. (2), the free-space monopolar Green’s function is chosen as
the replica vector since we expect to retrieve a local source for which
the geophone array is located at one or two wavelengths away from
the source. In this case, the separation between Rayleigh waves and
body waves is not effective and wave propagation can be modelled
by a velocity c that depends on the medium physical properties.
Because of the simple form of the replica vector in eq. (2), MFP
could also be described as spherical beamforming. However, more
complex Green’s function could be used as replica vectors in the
case of a forward model with layering, for example.
The linear MFP (Bartlett) processor is estimated as follows:
BBart (ai ) =
L
|dm ∗ ( f j , ai ) · K ( f j ) · dm ( f j , ai )|.
(3)
j=1
Similarly, the non-linear processor (MVDR) output is formulated
as
L 1
(4)
B M V (ai ) =
d ∗ ( f , a ) · K −1 ( f ) · d ( f , a ) .
m
j
i
j
m
j
i
j=1
As shown in eqs (3) and (4) above, the MFP is typically averaged
incoherently over a set of frequencies f 1 , f 2 , . . . , fL to improve the
contrast of the MFP output.
However, the MFP can be processed coherently by considering
the cross-correlation field instead of the acoustic noise data to construct the cross-spectral density matrix K. The coherent use of MFP
C 2011 The Authors, GJI, 187, 385–393
C 2011 RAS
Geophysical Journal International Old Faithful viewed through array processing
389
where p is the number of correlation functions selected among the
geophone array. The CSDM is calculated as before,
(6)
K̂ = dˆ · dˆ∗ .
Since the data are now issued from correlations between sensor
pairs at different frequencies, the replica vectors have to follow the
same logic. This means that the replica vector is expressed by
1
−2πi f (ai − aref )
,
(7)
exp
dˆm ( f, ai , aref ) =
16π 2 ai aref
c
where ai and aref refer now to the distance between the candidate
source position and, respectively, the ith geophone or the reference
geophone. The model-based replica is then compiled into a ‘supervector’ dˆm equivalent to the data ‘supervector’ from which the
linear and non-linear coherent MFP are computed as
ˆ (ai ) = |dˆ∗m (ai ) · K̂ · dˆm (ai )|;
(8)
BBart
1
ˆ
B M V (ai ) = .
dˆ∗m (ai ) · K̂ −1 · dˆm (ai ) (9)
It has been shown that the coherent MFP yields better results
than the incoherent approach for tracking objects in the ocean
(Michalopoulou & Porter 1996; Debever & Kuperman 2007). The
disadvantage of coherent processing is that it requires the manipulation of large matrices which may considerately increase the computation time.
Moreover, a fundamental requirement for MFP processing is that
the signal recorded at the sensors is coherent from one geophone
to another. This often limits its application to low frequency as
will be shown in the next section. A good first guess is required of
the medium velocity, especially for the MVDR where small speed
mismatch can degrade the resolution of the source localization. Two
standards are used to evaluate the MFP result: (1) focalization (size
of the focal spot) and (2) contrast (ratio between the maximum of
the MFP output and eventual sidelobes).
Figure 4. Cross-correlation of 10 min of seismic signal recorded on the
96 geophones with the sensor 54 as reference. The signals are bandpass
filtered between 10 and 14 Hz. The symbols correspond to theoretical delays
associated with the point source retrieved with MFP (X = −1.35 m, Y =
1.65 m and Z = 11 m) in a 1-D model with a vertical gradient of velocity of
23.5 m s−1 m−1 and a surface velocity of 130 m s−1 , the black stars indicate
the sensors used for the processing and the magenta circles are associated
to sensors that were disregarded in the MFP processing.
implies a coherent average over a discrete number of frequencies
in the bandwidth of interest, which requires the source signal to be
isolated in the data. This is done by cross correlating the noise signal recorded on each element of the array to a reference geophone
(Fig. 4). The coherent approach can be used in two ways. The first
considers correlations associated with one reference geophone only,
then separately calculates the MFP using the correlations with different geophones and averages the different MFP outputs. A better
approach consists in (1) calculating all correlations between the geophones and (2) selecting a set of p correlations that correspond to an
homogeneous distribution of station pairs among the network, that
is, the interstation paths cover uniformly the whole area (Fig. 8c).
To consider a coherent MFP processing, the set of correlations are
transformed into the frequency domain as data vectors at frequencies
ˆ
f 1 , f 2 , . . . , fL . We then create a ‘supervector’ d.
p
p
dˆ = [d 1f1 , d 2f1 , . . . , d f1 , . . . , d 1f L , . . . d f L ],
C
2011 The Authors, GJI, 187, 385–393
C 2011 RAS
Geophysical Journal International (5)
3.2 Processing
The first step in the MFP processing is the selection of the appropriate frequency bandwidth. A few points have to be taken into
considerations. First, the higher the frequency, the shorter the wavelength and the better the spatial resolution of the MFP localization.
MFP is based on the spatial coherence of the recorded signals, which
tends to decrease at higher frequencies and limits the use of large
arrays. Furthermore, MFP always applies a comparison between
the data and a wave-propagation model. The higher the frequency,
the more complicated the model must be since short wavelengths
are typically more sensitive to spatial heterogeneities. Finally, the
propagation model used for MFP in the case of a broad frequency
bandwidth must also include a frequency-dependent velocity profile. A balance between MFP resolution at high frequencies and
robust MFP localization at low frequencies is problem-specific and
must be determined on a case by case basis.
In the case of Old Faithful data, the spatial coherence was first
calculated from 8 to 70 Hz. The coherency is high between 12
and 58 Hz, while the signals are most energetic in the frequency
band 10–40 Hz (Fig. 3b). Finally, comparing the MFP results in the
5–15 Hz and 20–30 Hz bands, it appeared that the focalization and
the contrast are better in the lower frequency band. The MFP was
then processed between 5 Hz and 15 Hz with a 1 Hz sliding frequency window, and the contrast and the focalization were optimal
at 12 Hz.
390
E. Cros et al.
In the final step, we selected the sensors to be used with the
incoherent processing or the sensor pairs in the case of coherent
processing. Sensors located in a lower velocity zone and showing
a degraded spatial coherence were rejected (Fig. 4). For coherent
MFP processing, we restricted our choice to five reference stations
among the network to (1) provide homogeneous spatial distribution
of the station pairs while (2) limiting the size of the K̂ and thus the
computation time of the MFP processing.
3.3 Results and discussion
Figure 5. A velocity model proposed by Kedar (1996) compared to the
velocity model used for the localization.
In the second step, an estimation of the seismic velocity was
performed using the records of 12 hammer shots made by Kedar
in 1992. This analysis revealed that the mean surface velocity is
∼130 m s−1 between 11 and 13 Hz, with a low-velocity area in the
South part of the network. This zone of lower velocity could be
associated with softer sediments deposited in a small stream area.
When performing MFP, we have used this mean surface velocity
(130 m s−1 ) in a 1-D tabular model. The vertical gradient was estimated from a velocity model of Kedar (1996) using shear wave
velocity model, as shown in Fig. 5 and is of 23.5 m s−1 m−1 .
The 10 min of recorded signal was processed to localize and monitor
the dominant noise source position. The signals were truncated into
chunks of T = 20 s time window from which coherent/incoherent
MFP was performed using either the linear Bartlett or non-linear
MVDR method.
Figs 6 and 7 shows incoherent MFP results for one T = 20 s time
window using Bartlett and MVDR processing. The MFP results
are displayed as 3-D maps that correspond to the probability of
presence of the noise source (Figs 6a and 7a). We first note that
both linear/non-linear MFP give the same general source position.
The spatial resolution of the MFP is evaluated from slices in the
X –Y , X –Z and Y –Z planes at the MFP maximum (Figs 6b–d and
7b–d). As expected, the incoherent MVDR performs better than the
Bartlett in terms of spatial focalization. Indeed, the spatial resolution
of the linear Bartlett MFP is limited to the half-wavelength (∼6.5 m)
according to diffraction laws while the non-linear MVDR MFP
outpasses this limit with an ∼2 m spatial resolution.
When compared to incoherent MFP, coherent MFP does not improve the focalization results as shown in Figs 8(a) and (b). Compared to ocean where coherent processing significantly improved
the focalization performance (Debever & Kuperman 2007), the optimal focalization limit was already reached with incoherent MFP
in this case thanks to the high signal-to-noise ratio of the seismic
signals and the dense spatial coverage provided by the geophone
array around the geyser vent.
To confirm the validity of the MFP results, traveltimes were calculated between the MFP source position and each sensor according to
the velocity model plotted in Fig. 5. When compared to a reference
Figure 6. (a) 3-D Incoherent Bartlett output between 11.5 and 12.5 Hz for a medium with a velocity of 130 m s−1 at surface and a gradient of velocity of
23.5 m s−1 m−1 , with (b) slice in the plane X –Z, (c) slice in the plane Y –Z and (d) slice in the plane X –Y . The white circle corresponds to the horizontal vent
location. The source was determined using the search grid represented on Fig. 2(b).
C 2011 The Authors, GJI, 187, 385–393
C 2011 RAS
Geophysical Journal International Old Faithful viewed through array processing
391
Figure 7. (a) 3-D Incoherent MVDR output between 11.5 and 12.5 Hz for a medium with a velocity of 130 m s−1 at surface and a gradient of velocity of
23.5 m s−1 m−1 , with (b) slice in the plane X –Z, (c) slice in the plane Y –Z and (d) slice in the plane X –Y . The white circle corresponds to the horizontal vent
location. The source was determined using the search grid represented in red on Fig. 2(b).
Figure 8. Estimated locations of the seismic sources using the whole 10 min of signal according to the MFP method. Each error bar refers to the spot width
measured at 70 per cent of the maximum. The methods used are B: Bartlett and M: MVDR. These methods were processed (a) incoherently, (b) coherently
with correlations calculated between 11.5 and 12.5 Hz from p = 171 station pairs. The dotted lines on grey represent the position of the vent at the surface. (c)
Number of paths per cell projected on a 70 × 70 m grid with 5-m squared cells around the geyser position for the stations pairs selected for the coherent MFP.
geophone, these theoretical time delays were then superimposed to
the cross-correlation function with the same reference geophone
(Fig. 4). The satisfactory adjustment of the theoretical time delays
with the dominant cross-correlation wave front over most of the
geophone array is an a posteriori validation of MFP results. The
discrepancy observed for sensors 17–20 may be due to wrong station coordinates or to the presence of a strong spatial heterogeneity
in the medium.
C
2011 The Authors, GJI, 187, 385–393
C 2011 RAS
Geophysical Journal International The 12-m depth of the noise source is consistent with in situ
observations (Hutchinson et al. 1997). The temperature measurements made by Birch & Kennedy (1972) indicate a stationarity of
the water level in the conduit during the same cycle period, which
was confirmed by Hutchinson et al. (1997). Furthermore, in situ
observations with a camera made by Hutchinson revealed the presence of a widening of the conduit between 10.5 and 14 m. The
horizontal location of the source with the different MFP processors
392
E. Cros et al.
Figure 9. (a) Location of the seismic sources in a X –Y plane during 10 min of signal determined with coherent Bartlett processed on 20-s-long windows and
with an overlap of 75 per cent. The position of the source is relative to the position of the vent. (b) Location of the seismic sources on a X –Z plane and (c) on a
Y –Z plane. (d) Location of the seismic source in depth according to the time. (e) Fourier Transform of the source position dynamics averaged on the X , Y and
Z directions.
closely corresponds to the orifice position (Figs 8a and b). The digression of the source position from the horizontal location of the
vent maybe due to the widening of the conduit at depth. This shift
could also be attributed to the uncertainty on the velocity model and
the geophone positions.
The monitoring of the noise source inside the vent was performed
for each successive T = 20 s time window with an overlap of 75
per cent. The spatial localization of the noise sources is shown in
Figs 9(a)–(c), showing stable results during the 10-min-long recording. The standard deviation of the source position in the X and Y
direction is 0.30 m, while standard deviation is slightly larger in
the Z direction with a value of 0.42 m. More precisely, the source
depth shows periodic variation with a dominant period slightly less
than 1 min (Figs 9d and e). This period may be associated with temperature oscillation observed at this depth (Hutchinson et al. 1997),
likely due to two-phase flow static instabilities (Bouré et al. 1973).
4 C O N C LU S I O N
The efficiency of the MFP method was demonstrated in retrieving the location of the dominant noise source in hydrothermal
systems.
C 2011 The Authors, GJI, 187, 385–393
C 2011 RAS
Geophysical Journal International Old Faithful viewed through array processing
Using a velocity model (Kedar 1996) and the mean surface velocity calculated using hammer shots, the origin of the seismic signals
recorded at OFG on a geophone network at the surface is in good
agreement with in situ measurements. The dominant seismic source
location during this period of record corresponds to a steady state
with continuous boiling rate at a given constant depth (∼12 m).
The data processing using different MFP techniques show similar
source locations. Differences in the MFP results concern the spatial
width of the focalization according to the MFP technique. The
MVDR MFP proved to provide higher resolution results than the
Bartlett MFP for all cases analysed in this study, resulting in an
∼2 m source resolution for the MVDR MFP, compared to ∼4 m
resolution for Bartlett MFP.
The time evolution of the source location of the multiple impulsive events was continuously followed during a 10-min-long steady
period of seismic activity and showed a stable source position, with
fluctuations of small amplitude (less than 50 cm) and a period less
than 1 min. Applying MFP techniques on longer data set, comprising several cycles, would permit to perform a temporal monitoring
of the acoustic source and to improve the understanding of the
geyser dynamics.
Furthermore, the MFP method could be an interesting tool to
monitor other volcanic signals like volcano-tectonic event (VT) or
an LP event.
AC K N OW L E D G M E N T S
Most of the computations presented in this paper were performed
at the Service Commun de Calcul Intensif de l’Observatoire de
Grenoble (SCCI). The Digital Elevation Model used in this study
is based on data services provided by the OpenTopography Facility with support from the National Science Foundation under NSF
Award Numbers 0930731 and 0930643 and is based on services provided by the Plate Boundary Observatory operated by UNAVCO for
EarthScope (http://www.earthscope.org) and supported by the National Science Foundation (No. EAR-0350028 and EAR-0732947).
We would like to thank Shaul Hurwitz and Robert Clayton for their
help.
REFERENCES
Aki, K. & Ferrazzini, V., 2000. Seismic monitoring and modeling of
an active volcano for prediction, J. geophys. Res., 105(B7), 16 617–
16 640.
Almendros, J., Chouet, B. & Dawson, P., 2001. Spatial extent of a hydrothermal system at Kilauea Volcano, Hawaii, determined from array analyses
of shallow long-period seismicity. 1. Method, J. geophys. Res., 106(B7),
13 565–13 580.
Battaglia, J. & Aki, K., 2003. Location of seismic events and eruptive fissures
on the Piton de la Fournaise Volcano using seismic amplitudes, J. geophys.
Res., 108(B8), 2364, doi:10.1029/2002JB002193.
Battaglia, J., Got, J.-L. & Okubo, P., 2003. Location of long-period events
below Kilauea Volcano using seismic amplitudes and accurate relative
relocation, J. geophys. Res., 108(B12), 2553, doi:10.1029/2003JB002517.
Birch, F. & Kennedy, G.C., 1972. Notes on geyser temperatures in Iceland
and Yellowstone National Park, in Flow and Fracture of Rocks, Vol.16,
pp. 329–336, American Geophysical Union Geophysical Monograph.
Bouré, J., Bergles, A. & Tong, L., 1973. Review of two-phase flow instability,
Nucl. Eng. Des., 25(2), 165–192.
De Barros, L., Bean, C.J., Lokmer, I., Saccorotti, G., Zuccarello, L.,
O’Brien, G.S., Métaxian, J.-P. & Patanè, D., 2009. Source geometry
from exceptionally high resolution long period event observations at Mt
C
2011 The Authors, GJI, 187, 385–393
C 2011 RAS
Geophysical Journal International 393
Etna during the 2008 eruption, Geophys. Res. Lett., 36(24), L24305,
doi:10.1029/2009GL041273.
Debever, C. & Kuperman, W.A., 2007. Robust matched-field processing
using a coherent broadband white noise constraint processor, J. acoust.
Soc. Am., 122(4), 1979–1986.
Fournier, R.O., 1989. Geochemistry and dynamics of the Yellowstone National Park hydrothermal system, Annu. Rev. Earth planet. Sci., 17(1),
13–53.
Furumoto, M., Kunitomo, T., Inoue, H., Yamada, I., Yamaoka, K., Ikami, A.
& Fukao, Y., 1990. Twin sources of high-frequency volcanic tremor of
Izu-Oshima Volcano, Japan, Geophys. Res. Lett., 17(1), 25–27.
Got, J.-L., Fréchet, J. & Klein, F.W., 1994. Deep fault plane geometry inferred
from multiplet relative relocation beneath the south flank of Kilauea, J.
geophys. Res., 99(B8), 15 375–15 386.
Hurwitz, S., Kumar, A., Taylor, R. & Heasler, H., 2008. Climate-induced
variations of geyser periodicity in Yellowstone National Park, USA, Geology, 36(6), 451–454.
Husen, S., Smith, R.B. & Waite, G.P., 2004. Evidence for gas and magmatic
sources beneath the Yellowstone volcanic field from seismic tomographic
imaging, J. Volc. Geotherm. Res., 131(3-4), 397–410.
Hutchinson, R.A., Westphal, J.A. & Kieffer, S.W., 1997. In situ observations
of Old Faithful Geyser, Geology, 25(10), 875–878.
Ichihara, M. & Nishimura, T., 2011. Pressure impulses generated by bubbles
interacting with ambient perturbation, in Extreme Environmental Events,
pp. 731–752, Springer, New York, NY.
Jensen, F.B., Kuperman, W.A., Porter, M.B., Schmidt, H. & Bartram, J.F.,
1995. Computational Ocean Acoustics, American Institute of Physics,
New York, 578 pp.
Kedar, S., 1996. The origin of harmonic tremor at Old Faithful, PhD thesis,
Caltech, Pasadena, CA.
Kedar, S., Sturtevant, B. & Kanamori, H., 1996. The origin of harmonic
tremor at Old Faithful Geyser, Nature, 379, 708–711.
Kedar, S., Kanamori, H. & Sturtevant, B., 1998. Bubble collapse as
the source of tremor at Old Faithful Geyser, J. geophys. Res., 103,
24 283–24 299.
Kieffer, S.W., 1984. Seismicity at Old Faithful Geyser: an isolated source
of geothermal noise and possible analogue of volcanic seismicity, J. Volc.
Geotherm. Res., 22(1-2), 59–95, doi:10.1016/0377-0273(84)90035-0.
La Rocca, M., Galluzzo, D., Malone, S., McCausland, W. & Del Pezzo,
E., 2010. Array analysis and precise source location of deep tremor in
Cascadia, J. geophys. Res., 115, B00A20, doi:10.1029/2008JB006041.
Legaz, A., Revil, A., Roux, P., Vandemeulebrouck, J., Gouédard, P., Hurst,
T. & Bolève, A., 2009. Self-potential and passive seismic monitoring of
hydrothermal activity: a case study at Iodine Pool, Waimangu geothermal
valley, New Zealand, J. Volc. Geotherm. Res., 179(1-2), 11–18.
Métaxian, J.-P., Lesage, P. & Valette, B., 2002. Locating sources of volcanic tremor and emergent events by seismic triangulation: application to Arenal volcano, Costa Rica, J. geophys. Res., 107(B10), 2243,
doi:10.1029/2001JB000559.
Michalopoulou, Z.-H. & Porter, M.B., 1996. Matched-field processing for
broad-band source localization, IEEE J. Ocean. Eng., 21, 384–392.
Miller, D.S. & Smith, R.B., 1999. P and S velocity structure of the Yellowstone volcanic field from local earthquake and controlled-source tomography, J. geophys. Res., 104(B7), 15 105–15 121.
O’Brien, G.S., Lokmer, I., De Barros, L., Bean, C.J., Saccorotti, G.,
Métaxian, J.-P. & Patanè, D., 2011. Time reverse location of seismic longperiod events recorded on Mt Etna, Geophys. J. Int., 184(1), 452–462.
Rinehart, J., 1969. Old Faithful Geyser performance 1870 through 1966,
Bull. Volcanol., 33(1), 153–163.
Thode, A.M., D’Spain, G.L. & Kuperman, W.A., 2000. Matched-field processing, geoacoustic inversion, and source signature recovery of blue
whale vocalizations, J. acoust. Soc. Am., 107(3), 1286–1300.
Vandemeulebrouck, J., Roux, P., Gouédard, P., Legaz, A., Revil, A., Hurst,
A., Bolève, A. & Jardani, A., 2010. Application of acoustic noise
and self-potential localization techniques to a buried hydrothermal vent
(Waimangu Old Geyser site, New Zealand), Geophys. J. Int., 180(2),
883–890.